From Theory to Reality: Applying Conditional Burn-In to Drive Innovation in Industry
Considering a pre-defined quality target, it is state-of-the-art to assess optimal BI requirements based on a random sample of devices put to BI, a so-called BI study. Currently, it is assumed that the early life failure probability is constant for each technology. However, in reality, strongly depends on the production process. Thus, modelling dependent on the manufacturing enables a specification of individual BI requirements, which implies an overall reduction of BI time and costs.
In the iRel40 Project, Infineon Technologies Austria AG developed a new concept (Figure 1), that
- applies AI modes to infer a lot-specific health indicator on the basis of production data from different sources,
- links via a logistic regression model with the early life failure probability,
- applies a novel Clopper-Pearson-like interval estimator on the BI study data to estimate p.
Figure 1: Concept for lot-specific BI.
In the last months, IFAT could verify the overall concept based on simulated data. Moreover, the business case of this concept was successfully evaluated. Consequently, the new BI approach will be implemented within Infineon in the next years. This is a further step to drive innovation in the industry.
Dr. Horst Lewitschnig
Dr. Konstantin Posch
Burn-In, Reliability, Artificial Intelligence